Performance Comparison Between a Statistical Model, a Deterministic Model, and an Artificial Neural Network Model for Predicting Damage From Pitting Corrosion

Various attempts have been made to develop models for predicting the development of damage in metals and alloys due to pitting corrosion. These models may be divided into two classes: the empirical approach which employs extreme value statistics, and the deterministic approach based on perceived mechanisms for nucleation and growth of damage. More recently, Artificial Neural Networks (ANNs), a nondeterministic type of model, has been developed to describe the progression of damage due to pitting corrosion. We compare the three approaches above-statistical, deterministic, and neural networks. Our goal is to illustrate the advantages and disadvantages of each approach, in order that the most reliable methods may be employed in future algorithms for predicting pitting damage functions for engineering structures. To illustrale the difficulty that we face in predicting cumulative pitting damage, we selected a set of data that was collected in the laboratory. We compare and contrast the three approaches by reference to this data set.